GWAS-biologically informed SNP selection improves genomic relationship matrices and genetic parameter estimation in forest trees: insights from spruce and Eucalyptus

Accurate estimation of quantitative genetic parameters is critical for genomic selection in forest tree breeding. Compared with pedigree-based relationship matrices ( A -matrices), genomic relationship matrices ( G -matrices) better capture realized genetic relationships, including Mendelian sampling and cryptic relatedness. This study evaluated a genome-wide association study (GWAS)-based SNP selection strategy to improve G -matrix construction and estimates of additive genetic variance, heritability, and breeding values in white spruce (Picea glauca) and Eucalyptus (Eucalyptus grandis), two species with contrasting genome sizes and genotyping platforms. Progeny trials were analyzed using both pedigree-based best linear unbiased prediction (ABLUP) and genomic BLUP (GBLUP). Genomic relationship matrices were constructed from complete SNP datasets and compared with pedigree-based relationship matrices. To evaluate the impact of marker selection, SNPs were ranked according to the absolute magnitude of their GWAS-estimated effects, and sequential subsets of informative markers were used to reconstruct G -matrices. Parameter estimates and theoretical accuracy from reduced SNP panels were compared with those from the complete datasets. Genomic relationship matrices generated connected family networks, revealing hidden relatedness and overcoming limitations of pedigree-based relationships. GBLUP yielded consistently higher heritability estimates for all traits in white spruce but only for tree volume in Eucalyptus. GWAS-informed SNP selection showed that marker informativeness was more important than marker number; in white spruce, ~50,000 informative SNPs performed comparably to the full 467K SNP panel for some traits, although marker requirements varied among traits. In contrast, gains in Eucalyptus were limited, likely due to reduced representation of trait-associated variation. The results demonstrate that integrating realized genomic relationship matrices with biologically informed SNP selection can provide genetic parameter estimates and theoretical accuracies comparable to those obtained with much larger marker sets while substantially reducing genotyping requirements for genomic evaluation. Responses were species- and trait-dependent. Reduced gains in Eucalyptus may reflect SNP ascertainment bias and other factors limiting the representation of causal genetic variation. Overall, biologically relevant and genomically representative markers were more important than marker density alone, and optimized SNP selection may reduce genotyping requirements while maintaining effective genomic evaluation in forest tree breeding.